Abstract: Accurate prediction of the development height of water-conducting fractured zones is directly related to the safe production of coal mines and remains a key and difficult issue to be addressed in the field of mining engineering. Currently, the BP neural network, as a relatively mature algorithm, is widely used in predicting the height of water-conducting fractured zones. However, it has drawbacks such as unreasonable weight distribution and a tendency to fall into local optima, resulting in prediction accuracy that fails to meet the needs of actual production. To improve the prediction accuracy of the BP neural network, this study takes the Huainan Mining Area as the research region, collects 49 sets of measured data on the height of water-conducting fractured zones, and selects mining thickness, coal seam dip angle, mining depth, working face slope length, and hard rock proportion coefficient as influencing factors. The Grey Wolf Optimizer （GWO） is introduced to optimize the weights and thresholds of the BP neural network, thereby constructing a GWO-BP neural network model for predicting the height of water-conducting fractured zones. A total of 44 sets of data are used as training samples, and 5 sets as test samples to verify the prediction performance of the model, which is further compared with the prediction results of the BP neural network and empirical formulas. The results show that the BP neural network optimized by GWO exhibits significantly improved prediction performance, with higher accuracy and stability than both the unoptimized BP neural network and empirical formulas. For the test samples, the mean absolute error does not exceed 0.51 m, and the mean relative error is within 1.12%. The trained model is applied to the 1312（1） working face in the North No.2 Mining Area of Gubei Coal Mine. A comparison between the predicted results and the traditional measured data from surface boreholes, as well as the comprehensive measured data from distributed optical fiber and high-density electrical method, shows an error of 0.26 m and a relative error of 0.54%, respectively. These findings indicate that the GWO-BP neural network has good engineering applicability and can provide reliable technical support for water prevention and control work in mine roof areas.